EUROSPEECH 2001 Scandinavia
Line Spectrum Frequencies (LSFs) are efficient and popular for representing the spectral envelope in low bit-rate (LBR) speech coding. It is also attractive to use LSFs in the task of speech or speaker recognition. Although, the LBR speech coding does not deteriorate the recognition performance substantially, additive noise does degrade the performance. This paper presents an algorithm for finding LSFs of the addition of two speech signals. This algorithm can be used to adapt the model of noisy speech in LSFs domain, thereby improve the robustness of speech recognition in noisy environments. The experiments on Mandarin digits recognition have proved the effectiveness of the proposed algorithm.
Bibliographic reference. Yu, An-Tze / Wang, Hsiao-Chuan (2001): "An algorithm for finding line spectrum frequencies of added speech signals and its application to robust speech recognition", In EUROSPEECH-2001, 2657-2660.